3D eye model-based gaze estimation from a depth sensor

摘要: In this paper, we address the 3D eye gaze estimation problem using a low-cost, simple-setup, and non-intrusive consumer depth sensor (Kinect sensor). We present an effective accurate method based on model to estimate point of subject with tolerance free head movement. To determine parameters involved in proposed model, propose i) improved convolution-based means gradients iris center localization accurately efficiently locate space; ii) geometric constraints-based eyeball under constraints that all points are distributed sphere originated from sizes two eyeballs identical; iii) Kappa angle calculation fact visual axes both eyes intersect at same screen plane. The final is calculated by estimated parameters. experimentally evaluate our five subjects. experimental results show good performance average accuracy 3.78°, which outperforms several state-of-the-arts.